In the mining industry, a framework exists for quantitative assessment of interpretation uncertainty of spatial domains used to model a stationary spatial domain for mineral resource estimation. This framework will improve public reporting of mineral resource estimates, and improve there liability of feasibility studies by ensuring successful communication of geological risk. In early-stage mineral projects,there is often not enough multi element laboratory data to enable the use of calculated geological methods for quantitative uncertainty assessment. Portable X-Ray Fluoresce (pXRF) is an accepted method of providing cost and time effective multielement measurements for early-stage projects. However, these measurements are of lower precision and accuracy, then laboratory-based measurements. Recent work has shown that quantitative uncertainty assessments using a Bayesian approximation method can successfully use both pXRF and laboratory data. Subjective visual assessment of uncertainty band graphs, drill hole plots, and confidence matrices suggest that models derived from the two types of data provide similar uncertainty assessments. This paper reviews recent advances in Null Hypothesis and Bayesian Hypothesis statistical methods for comparing models to propose a robust methodological framework for assessing the reliability and similarity of supervised classification models utilising confusion matrix model metrics for further research in the use of pXRF as a suitable measurement for geological spatial domain uncertainty.
Original languageEnglish
Title of host publication2020 Higher Degree Research (HDR) Conference
EditorsMichelle Gilmour
PublisherCharles Sturt University
Number of pages4
Publication statusPublished - 21 Oct 2020
Event2020 Higher Degree Research (HDR) Online Conference: School of Computing and Mathematics Faculty of Business, Justice and Behavioural Sciences - Online
Duration: 21 Oct 202022 Oct 2020


Conference2020 Higher Degree Research (HDR) Online Conference
Internet address


Dive into the research topics of 'A robust framework for comparing classification models in spatial domains'. Together they form a unique fingerprint.

Cite this